Towards a Method to Detect F-formations in Real-Time to Enable Social Robots to Join Groups

Abstract

In this paper, we extend an algorithm to detect constraint based F-formations for a telepresence robot and also consider the situation when the robot is in motion. The proposed algorithm is computationally inexpensive, uses an egocentric (first-person) vision, low memory, low quality vision settings and also works in real time which is explicitly designed for a mobile robot. The proposed approach is a first step advancing in the direction of automatically detecting F-formations for the robotics community.